experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity
88,838 Commits 49 Branches 133 Tags 692 MiB
snyk-fix-f4c840df4db9b84b92bab0ff21780e61
Commit Graph

872 Commits

Author SHA1 Message Date
Frank Chen
23e387975e Rename tpu_load_library to tpu_api_dlsym_initializer. 
This better reflects what the load library module does.

PiperOrigin-RevId: 317763973
Change-Id: I66169394c2f1d77b260323cb81d76f0f4e167579
 2020-06-22 17:45:58 -07:00
TensorFlower Gardener
7c38468051 Merge pull request from nluehr:TF32-v2 
PiperOrigin-RevId: 317757557
Change-Id: I0a0f0cc9025db7d1fbc7975b07d7d934c6fa8c2f
 2020-06-22 17:23:41 -07:00
Nathan Luehr
3f2c98610e Move ScopedCublasMathMode inside DoBlasInternalImpl 2020-06-22 10:53:03 -07:00
Nathan Luehr
9afaf559d9 Locate CUDNN_FMA_MATH inside CUDNN 8 macros 2020-06-22 12:28:02 -05:00
Wenhao Jia
715b02167d Restore TpuPlatform auto registration code. 
PiperOrigin-RevId: 317409587
Change-Id: If44d7a39a45c4c7026f70a4d79d965a54c4db295
 2020-06-19 17:31:38 -07:00
Nathan Luehr
54a7bd1d73 Remove unnecessary tf32_execution_allowed check 2020-06-19 18:02:40 -05:00
Nathan Luehr
434f49d54f Rework gemm TF32 2020-06-19 15:12:30 -07:00
Nathan Luehr
6ff37474dd Use CUDNN_TENSOR_OP_MATH to enable tensor cores. 2020-06-19 15:12:30 -07:00
Nathan Luehr
03d836e261 Plumb TF32 for cublas gemm 2020-06-19 15:12:30 -07:00
Nathan Luehr
107e634823 Plumb TF32 for RNN 2020-06-19 15:12:30 -07:00
Nathan Luehr
376efd71b1 Convolution TF32 Plumbing 2020-06-19 15:12:30 -07:00
Nathan Luehr
32d63d0a3e Removed TENSOR_OP disable env vars. 
* TF_DISABLE_CUBLAS_TENSOR_OP_MATH
* TF_DISABLE_CUDNN_TENSOR_OP_MATH
* TF_DISABLE_CUDNN_RNN_TENSOR_OP_MATH
 2020-06-19 15:12:30 -07:00
A. Unique TensorFlower
9c4b749b09 Internal change 
PiperOrigin-RevId: 317230321
Change-Id: I043dae37768f6e9cf946d4db2a8c36123ed2d6d9
 2020-06-18 19:24:46 -07:00
TensorFlower Gardener
18ca7c95d4 Merge pull request from nluehr:TF32 
PiperOrigin-RevId: 317221015
Change-Id: I228bfa51b797fe52112012932a1dba9a356151ba
 2020-06-18 18:09:37 -07:00
A. Unique TensorFlower
aab151356d [pjrt] Refresh stream error status in strategic places to flush out silent failures. 
PiperOrigin-RevId: 317204018
Change-Id: If75a3ad9ec846ce1621cdba92a2dc738b65b7001
 2020-06-18 16:28:27 -07:00
Frank Chen
cf00e559d7 Add TPU configuration ops to default TensorFlow build 
PiperOrigin-RevId: 317133514
Change-Id: I33bc6d7fdbba5915bd0d1291d4e086139c07eb14
 2020-06-18 10:52:49 -07:00
Nathan Luehr
3ebc53c394 Relax stub include version checking. 
Remove upper bound on version check for latest inc files.
 2020-06-17 07:48:07 -07:00
A. Unique TensorFlower
267f956246 load libcupti.so shared object if libcupti.so.10.1 is not installed. This makes it possible for tensorflow to load CUPTI 10.2 via libcupti.so symlink in the event that libcupti.so.10.1 is missing. 
PiperOrigin-RevId: 316782424
Change-Id: I930912f1d3a8fa80e8b91c41214c374650f08847
 2020-06-16 16:49:55 -07:00
Christian Sigg
09cf51da59 Fix switch statement for CUDNN_CONVOLUTION_BWD_FILTER_ALGO_FFT_TILING. 
PiperOrigin-RevId: 316522188
Change-Id: I6557506f90d4bf6a91e97b65abd494d58769e3a3
 2020-06-15 12:43:25 -07:00
TensorFlower Gardener
4f07e60caf Merge pull request from nluehr:cudnn8_algo_cleanup 
PiperOrigin-RevId: 316162945
Change-Id: I72bc02421cbb545f68049a5f36c0a6aea7dbba1a
 2020-06-12 13:21:32 -07:00
Wenhao Jia
d7d97a0d72 Internal change only. 
PiperOrigin-RevId: 316058358
Change-Id: I2317fe006b40b9f930e6610ea7c059d15662dcb1
 2020-06-12 00:14:31 -07:00
Nathan Luehr
013ddd96d0 Use CUDNN_TENSOR_OP_MATH to enable tensor cores. 2020-06-11 16:41:28 -05:00
TensorFlower Gardener
f60069b7bb Merge pull request from nouiz:vlog_ptxas 
PiperOrigin-RevId: 315938190
Change-Id: Ib5ac12f94e3eaac65e97bdfd650af95bedf3f326
 2020-06-11 11:26:47 -07:00
Henry Tan
f418a84299 Reorganizing TPU OPS C_API related files. 
PiperOrigin-RevId: 315834750
Change-Id: I501ac67f75130e6a1114b75db5ea262e8ef3890c
 2020-06-10 22:22:48 -07:00
Henry Tan
c055650aba TPU internal change. 
PiperOrigin-RevId: 315644771
Change-Id: I43705573dd797add30f776ec5903c1f98c614061
 2020-06-10 00:48:07 -07:00
Nathan Luehr
9040800a7b Remove cudnn RNN algo APIs from cudnn 8 inc file 2020-06-09 11:41:59 -05:00
Frank Chen
13c09da422 Open sourcing some TPU-related work 
PiperOrigin-RevId: 315431095
Change-Id: I734632c0e5723dfca37acf53bbbd2b378b04c95d
 2020-06-09 00:03:31 -07:00
Frederic Bastien
3946ad2bfd Print the ptxas command line as higher VLOG level. 2020-06-08 13:57:12 -07:00
Frederic Bastien
ce806400a4 Print the verbose output of PTXAS. 2020-06-08 13:42:43 -07:00
TensorFlower Gardener
2b623518ef Merge pull request from ROCmSoftwarePlatform:google_upstream_rocm_csb_fix_200604_2 
PiperOrigin-RevId: 315326861
Change-Id: Idd32f299d5ad4b6cb89141b8888bdabb038cbea9
 2020-06-08 12:46:16 -07:00
TensorFlower Gardener
e63137a53c Merge pull request from bhack:patch-2 
PiperOrigin-RevId: 314873638
Change-Id: I9fa2b55613c78eb64dec1a40a1f157374498a929
 2020-06-04 23:22:55 -07:00
Deven Desai
856f899e4e [ROCm] Fix for the ROCm CSB breakage - 200604 
This PR addresses 2 unit test failures introduced by the following commit
bd20260350

failing unit tests

```
//tensorflow/core/common_runtime/gpu:gpu_device_test
//tensorflow/core/common_runtime/gpu:gpu_device_unified_memory_test
```

It is different from PR , which fixes the build error on the ROCm platform.

The commit above adds unit tests that check the assignment of priority values to GPU streams. Because ROCm support for assigning priority values to GPU streams is missing, those unit tests fail.

This PR/commit adds the missing ROCm support, and updates the unit test to work with AMD GPUs too. The valid priority value range seems to be different for AMD GPUs (0,2) as compared to NVidia GPUs (-1, 0), and hence the changes requrired in the testcases too.
 2020-06-05 00:36:29 +00:00
A. Unique TensorFlower
e9ad27a920 Add member typedef value_type to StatusOr to be conformant with the 
standard, e.g. `std::optional`, `std::expected`.

PiperOrigin-RevId: 314649934
Change-Id: Id9e9c70a486254db4221af550f0b3c77440b0fb6
 2020-06-03 19:25:02 -07:00
Nathan Luehr
b67608e66c Plumb TF32 for cublas gemm 2020-06-03 15:47:48 -05:00
Nathan Luehr
0f58bb6309 Plumb TF32 for RNN 2020-06-03 15:47:47 -05:00
Nathan Luehr
dedb51aec2 Convolution TF32 Plumbing 2020-06-03 15:47:47 -05:00
Nathan Luehr
ab6b2ffde3 Removed TENSOR_OP disable env vars. 
* TF_DISABLE_CUBLAS_TENSOR_OP_MATH
* TF_DISABLE_CUDNN_TENSOR_OP_MATH
* TF_DISABLE_CUDNN_RNN_TENSOR_OP_MATH
 2020-06-03 15:47:10 -05:00
Sreeni Kesavarapu
bd20260350 Implement GPU device priority for virtual devices. 
- This adds a field in virtual device options to specify priority for virtual devices.
  - When the priority is specified, it will be used to create the cuda streams for the virtual device with the given priority.
  - This is backwards compatible with no priorities specified. When no priorities specified, the current implementation of creating a stream without any priority will continue while any non-zero priorities specified will be used to create streams with that priority.

PiperOrigin-RevId: 314470276
Change-Id: I4943f71e901245fb21b6f7e833adbdcd8126f1fa
 2020-06-02 22:31:35 -07:00
Christian Sigg
255f590ab6 Roll PR (CUDNN v8 support) forward with fix: 
Handle '64_' prefix in the cuDNN version for Windows.

PiperOrigin-RevId: 314385732
Change-Id: I9b36497dbe460e2e7cad3d755cf9ce8d41647ce3
 2020-06-02 12:54:05 -07:00
Zhenyu Tan
5cf536f78d Rollback PR : CUDNN v8 support 
PiperOrigin-RevId: 314219496
Change-Id: I3e36453044e5b73274e05d84f8d7f2cdb5b144ae
 2020-06-01 15:57:18 -07:00
TensorFlower Gardener
8b4dd17f2b Merge pull request from nluehr:CUDNN8 
PiperOrigin-RevId: 313914508
Change-Id: Ib186d57303d220a49a0b27d8be591476e69dcfd2
 2020-05-30 01:18:08 -07:00
A. Unique TensorFlower
396105ddd5 fix build due to cupti static link. 
PiperOrigin-RevId: 313880914
Change-Id: I4cf691e60fd28e773f124e440ebb30f054d5f2e9
 2020-05-29 17:43:42 -07:00
A. Unique TensorFlower
8c63be3940 Internal change 
PiperOrigin-RevId: 313850352
Change-Id: I89584b0bcb4409eb74d21e31fb0eb68844186707
 2020-05-29 14:31:09 -07:00
A. Unique TensorFlower
13645c7fd2 use static linked cupti internally. 
PiperOrigin-RevId: 313822331
Change-Id: Ie2c1e4a2d92ed9f92e2dcd8d3bc354546ab4663f
 2020-05-29 13:47:38 -07:00
Sreeni Kesavarapu
27ac0a900e Implement GPU device priority for virtual devices. 
- This adds a field in virtual device options to specify priority for virtual devices.
  - When the priority is specified, it will be used to create the cuda streams for the virtual device with the given priority.
  - This is backwards compatible with no priorities specified. When no priorities specified, the current implementation of creating a stream without any priority will continue while any non-zero priorities specified will be used to create streams with that priority.

PiperOrigin-RevId: 313814564
Change-Id: Ie580ea3f8c7fc7248e40e1ae4c898f0a029fba35
 2020-05-29 13:15:24 -07:00
TensorFlower Gardener
4a9c6447f8 Merge pull request from nluehr:CUDA11 
PiperOrigin-RevId: 313758239
Change-Id: I962e4d81b91f2a1249a3749adb4579440059d6dc
 2020-05-29 11:57:26 -07:00
bhack
d38caa0466
add new line 2020-05-28 19:48:52 +02:00
bhack
0d687877e4
Change error level to warning 2020-05-28 19:00:57 +02:00
Christian Sigg
7b495df2da Add CUDA 11 stub files. 
PiperOrigin-RevId: 313335318
Change-Id: I6e7717b580540fa2b039057702115fd25c3c54ad
 2020-05-26 23:19:19 -07:00
Christian Sigg
65264bfc98 Format generated CUDA stub files. 
PiperOrigin-RevId: 313334624
Change-Id: Ie7fc91e0c59754b9d9a7db686f3876577c8164e7
 2020-05-26 23:10:31 -07:00